JPS59128682A - Character reader - Google Patents

Abstract

PURPOSE:To recognize a character easily and to improve the recognizing accuracy by using the similarity of an input character pattern to a reference character pattern in a recognition dictionary and the appearance frequency of a character or a subset at each character position on a business form to recognize the character. CONSTITUTION:A character written in the business form 1 is scanned by a scanning means 2 and converted into an electric signal, which is processed by binary-coding or the like in a preprocessing means 3. A feature extracting means 4 extracts the feature of a character pattern and a similarity calculating means 5 matches the patterns from the feature of the extracted character pattern and a reference character pattern previously stored in the recognition dictionary 6. The similarity is applied to a discriminating means 8 and a character corresponding to the character pattern is decided from the appearance frequency of a position on the business form 1 which is stored in an appearance frequency storing means 7 and is to be read out at present or the appearance frequency of a subset. The read-out result 9 is outputted and sent to an appearance frequency undating means 10 to update the appearance frequency.

Description

[Detailed Description of the Invention] [Technical Field of the Invention] In recognizing characters recorded on a form, etc., the present invention is based on the similarity between an input character/mother turn and a reference character Noreen in a recognition dictionary, and on a form. The present invention relates to a character reading device that recognizes and reads characters using the frequency of appearance of characters or subsets at each character position.

[Prior art]

Conventionally, this type of character reading device is configured to recognize and fail to read a character using only the degree of similarity between the four turns of an unknown character and a reference character pattern in a recognition dictionary. When recognizing multiple types of characters, including , there are five drawbacks: the character recognition rate decreases because there are many similar characters. A method has also been proposed in which a reading type or reading subset is specified for each reading field of a form, but this method has the disadvantage that it is extremely troublesome to create a form format.

[Summary of the invention]

This invention was made for the purpose of eliminating the drawbacks of the conventional ones as described above, and in forms other than text input, it is possible to distinguish between characters or subsets that are more likely to appear and those that are less likely to appear, depending on the character position of the form. From the viewpoint of the existence of characters or subsets, characters are recognized using the degree of similarity between the input character pattern and the reference character pattern in the recognition dictionary and the frequency of appearance of characters or subsets at each character position on the form. The object of the present invention is to provide a character reading device that can improve character recognition accuracy.

[Embodiments of the invention]

Examples of the present invention will be described below. The drawing shows a character reading device which is an embodiment of the present invention! It is a lock block diagram. In the figure, l is a form, 2 is a scanning means for scanning the characters on the form l and converting it into an electric signal, and 3 is before obtaining the binary value of the character from the electric signal of the character obtained by the scanning means 2. processing means; 4 is a feature extraction means for extracting features from the character e-turns preprocessed by the preprocessing means 3;
Reference numeral 6 denotes a recognition dictionary storing standard/lean features of characters to be recognized, and 5 a similarity measure for calculating similarity from the features extracted by the feature extraction means 4 and the features of the reference character/mother turn in the recognition dictionary 6. Calculation means, 7 is the appearance frequency of the character to be recognized at each character position on the form l, or the number of kanji, alphabetic characters, and numbers per month;
1 is an appearance frequency storage means for storing the appearance frequency of a subset of flat names, symbols, etc., 8 is a determination means, 9 is a reading result, 1
0 is an appearance frequency updating means that updates the appearance frequency of a character or a subset at each character position of the form 1 based on the character recognition result.

Next, the operation of the character reading device having the configuration shown in the above drawings will be explained. First, characters written on a form 1 are scanned by the scanning means 2 based on a predetermined form format and converted into electrical signals. The preprocessing means 3 obtains a character/4' turn from the electric signal of the character obtained by the scanning means 2, which has undergone preprocessing such as character binarization. The feature extraction means 4 extracts the features of the character pattern that has been preprocessed by the preprocessing means 3. Similarity calculation means 5
Now, the features of the input character/IP turn Pi obtained by the feature extracting means 4 are matched with the features of all the reference characters/mother turns Pj stored in the recognition dictionary 6 to determine the degree of similarity Sij. The appearance frequency storage means 7 stores, for each character to be recognized, the appearance frequency Fij at the position on the form 1 where the character to be currently read is written. The determining means 8 calculates f(81j,
Fij) is used to determine which character the input character Tsukturn Pi is, or whether to refuse to read it. Further, the reading result 9 obtained by the determining means 8 is sent to the appearance frequency updating means 10 to update the appearance frequency of the character. However, if the reading result 9 is a refusal to read, the reading result 9 is not sent to the appearance frequency updating means IO.

Now, as an actual concrete example, in normal cases, forms l having the same format are often read successively.

In this case, since the character appearance frequency is initially 0 for all characters, read the ID of form 1 (R separate bell), follow the form format based on it, and perform character recognition using only similarity. . Then, the appearance frequency of characters is updated based on the reading result 9. In this case, even if the reading result 9 is misread, the accuracy of character recognition at a practical level is 90 or more, so there is little negative effect on the appearance frequency of characters. As the appearance frequency of the characters is updated, the characters are recognized using the above-mentioned similarity and the appearance frequency of the characters. For example, when the katakana ``ri'' and the number ``7'' become candidate characters with the same degree of similarity for the character e-turn recorded at a certain character position in form 1, If the number "7" is overwhelmingly higher than the katakana "ri" in the appearance frequency of characters, the reading result 9 is set as the number "7". In other words, higher recognition accuracy can be obtained compared to the case where characters are recognized using only similarity. Here, when a form 1 with a different ID is received, character recognition is performed using the frequency of appearance of characters in the form format corresponding to the ID of the different form 1.

In addition, in the above embodiment, the case where the determination is made using the degree of similarity and the appearance frequency of characters was explained. Characters may be recognized using the appearance frequency of which subset is likely to appear among subsets such as , flat name, symbol, etc.

〔Effect of the invention〕

As explained above, in order to recognize characters recorded on a form etc., this invention uses the degree of similarity between the features extracted from input characters/grams and the features of reference characters/grams in the recognition dictionary. Since it is configured to recognize characters using the appearance frequency of each character position on a form or the appearance frequency of a subset, unlike conventional character reading devices of this type, characters cannot be recognized using only the degree of similarity. This has an excellent effect in that the accuracy of the recognition rate for characters can be further improved compared to the case where the method is used.

[Brief explanation of drawings]

The drawing is a block diagram showing a character reading device according to an embodiment of the present invention. In the figure, 1... form, 2... scanning means, 3...
・Pre-processing means, number: Feature extraction means, 5: Similarity calculation means, 6: Recognition dictionary, 7: Appearance frequency storage means, 8: Judgment means, 9: Reading As a result, lO・
...This is an appearance frequency update means. Agent Makoto Kuzuno

Claims (1)

[Claims] In a character reading device that recognizes and reads characters recorded on a form, etc., there is a scanning means that scans the characters according to a predetermined form format and converts them into electrical signals, and an electrical signal of the characters obtained by the scanning means. Preprocessing means for obtaining binary noso turns of characters, feature extraction means for extracting features from characters/mother turns preprocessed by the preprocessing means, and storing features of reference A' turns of characters to be recognized. recognition dictionary and the frequency of occurrence of the recognition target character at each character position on the form, or kanji, alphabetic characters, 1-number katakana,
an appearance frequency storage means for storing the appearance frequency of a subset of flat names, symbols, etc., and a similarity degree for calculating a degree of similarity from the features extracted by the feature extraction means and the features of the reference character noguturn in the recognition dictionary. comprising a calculation means and an appearance frequency updating means for updating the appearance frequency of a character or a subset at each character position of the form based on the character recognition result, and recognizing a character using the similarity and the appearance frequency. A character reading device characterized by a character reading of 5.